behavioral (only) analyses: master report
michael freund
2020-09-12
Generated from rmarkdown file: “code/behav/_behav.rmd“. This rmd file executes ‘child’ R-scripts within this directory. These child scripts contain analysis code. Their output is captured and ‘knitted’ back into this report. To run the code, you can either execute this .rmd file (i.e., ‘knit’ to html), or source each child script individually. (These child scripts should be configured to run on their own when sourced directly.)
1 analysis-set subjs: model construction, estimation of stroop effects
1.1 RT analysis
1.1.1 plot
plot(behav$rt)
abline(h = 250)
abline(h = 3000) ## marks beginning of subsequent trial
behav$is.implausible.rt <- behav$rt > 3000 | behav$rt < 250
grid.arrange(
behav %>%
filter(acc == 1, !is.implausible.rt) %>%
ggplot(aes(rt)) +
geom_density(fill = "slateblue", alpha = 0.3) +
labs(title = "all correct trials") +
theme(legend.position = "none"),
behav %>%
filter(acc == 1, !is.implausible.rt) %>%
ggplot(aes(rt)) +
geom_density(aes(fill = trial.type), alpha = 0.3) +
labs(title = "by congruency") +
scale_fill_brewer() +
scale_color_brewer() +
theme(legend.position = c(0.5, 0.5)),
ncol = 2
)
behav %>%
filter(acc == 1, !is.implausible.rt) %>%
full_join(group_by(., subj) %>% summarize(r2norm = qqr2(rt)), by = "subj") %>%
ggplot(aes(sample = rt)) +
stat_qq(alpha = 0.8, size = 1) +
stat_qq_line(size = 0.25) +
facet_wrap(vars(subj)) +
geom_text(
aes(
x = -1.25, y = 2000,
label = paste0("r^2 = ", round(r2norm, 3))
), size = 3, color = "grey50"
) +
theme_minimal(base_size = 10) +
theme(axis.title = element_blank()) +
labs(
title = "QQ rt versus normal",
subtitle = paste0("overall r^2 with normal = ", round(qqr2(behav$rt), 3))
)
849971 and 161832 have odd patterns in RT distribution. Strong clustering at 500 ms. This is highly consistent with an artifact we found in other RT data from the same microphone/preprocessing method.
1.1.2 handling likely-artifactual RTs
behav %>%
filter(acc == 1, !is.implausible.rt) %>%
ggplot(aes(x = rt, group = subj)) +
geom_density(data = . %>% filter(!subj %in% c("849971", "161832")), color = rgb(0, 0, 0, 0.15), size = 2) +
geom_density(data = . %>% filter(subj %in% c("849971", "161832")), color = "firebrick1", size = 2) +
labs(
title = "subjects with bimodal distributions (likely artifactual) in red"
)
behav$is.artifactual.rt <- FALSE
behav$is.artifactual.rt[behav$subj %in% c("849971", "161832") & behav$rt < 500] <- TRUE
behav %>%
filter(acc == 1, !is.implausible.rt, subj %in% c("849971", "161832")) %>%
full_join(group_by(., subj) %>% summarize(r2norm = qqr2(rt)), by = "subj") %>%
ggplot(aes(sample = rt)) +
stat_qq(alpha = 0.8, size = 0.4) +
stat_qq_line(size = 0.25) +
geom_hline(yintercept = 500) +
facet_wrap(vars(subj)) +
theme_minimal(base_size = 10)
1.1.3 fit RT model
## make sure to exclude validation set!
behav.rt.aset <- behav %>% filter(acc == 1, !is.implausible.rt, !is.artifactual.rt, is.analysis.group)
fit1.het <- lme(
rt ~ trial.type,
random = ~ trial.type | subj,
data = behav.rt.aset,
weights = varIdent(form = ~ 1 | subj),
control = cl1,
method = "REML"
)
## define "far" outliers as points with resids more extreme than 3 IQR
behav.rt.aset$resid.p <- resid(fit1.het, type = "p")
behav.rt.aset$is.far.out <- farout(behav.rt.aset$resid.p)
## exclude far outliers and refit model
fit1.het.trim <- update(fit1.het, data = behav.rt.aset %>% filter(!is.far.out))
fit1.het.trim.ml <- update(fit1.het.trim, method = "ML") ## for model comparisons1.1.4 test heterogeneity of level-I variance
fit1.hom.trim.ml <- lme(
rt ~ trial.type,
random = ~ trial.type | subj,
data = behav.rt.aset %>% filter(!is.far.out),
control = cl1,
method = "ML"
)
(rt.hom.v.het <- anova(fit1.hom.trim.ml, fit1.het.trim.ml))## Model df AIC BIC logLik Test L.Ratio p-value
## fit1.hom.trim.ml 1 6 132710.8 132754.1 -66349.38
## fit1.het.trim.ml 2 54 128508.2 128898.3 -64200.10 1 vs 2 4298.551 <.00011.1.5 test variance in stroop effect
fit0.het.trim.ml <- update(fit1.het.trim.ml, random = ~ 1 | subj)
(rt.stroopvar <- anova(fit0.het.trim.ml, fit1.het.trim.ml))## Model df AIC BIC logLik Test L.Ratio p-value
## fit0.het.trim.ml 1 52 128591.7 128967.4 -64243.87
## fit1.het.trim.ml 2 54 128508.2 128898.3 -64200.10 1 vs 2 87.53103 <.00011.1.6 estimate cross-run reliability
fit1.het.run.trim <- lme(
rt ~ trial.type * run,
random = ~ trial.type * run | subj,
data = behav.rt.aset %>% filter(!is.far.out),
weights = varIdent(form = ~ 1 | subj),
control = cl1,
method = "REML"
)
summary(fit1.het.run.trim)## Linear mixed-effects model fit by REML
## Data: behav.rt.aset %>% filter(!is.far.out)
## AIC BIC logLik
## 128353.5 128808.6 -64113.76
##
## Random effects:
## Formula: ~trial.type * run | subj
## Structure: General positive-definite, Log-Cholesky parametrization
## StdDev Corr
## (Intercept) 117.38521 (Intr) trl.ty run
## trial.typeincon 33.46107 0.008
## run 42.03440 0.315 -0.007
## trial.typeincon:run 11.09470 -0.430 0.084 0.708
## Residual 82.02205
##
## Variance function:
## Structure: Different standard deviations per stratum
## Formula: ~1 | subj
## Parameter estimates:
## 107321 123117 130114 132017 138837 141422
## 1.0000000 2.2042468 1.1374682 1.9329308 1.8295447 1.7368152
## 150423 158136 160830 161832 165032 171330
## 2.0390034 1.0513835 4.5850017 5.1846430 2.6913126 1.8198504
## 178243 178950 182436 197449 203418 204319
## 2.8558033 1.3055077 3.2531529 2.1067865 1.1163342 2.5114102
## 300618 317332 346945 352738 448347 580650
## 1.0329272 1.1376803 1.4761791 2.5163718 1.6665064 3.3350494
## 594156 601127 765864 814649 843151 849971
## 1.0921172 1.9365139 1.0036256 1.7891142 2.1310886 2.7519015
## 873968 877168 DMCC1971064 DMCC2442951 DMCC3062542 DMCC4854075
## 1.2534443 2.1359043 1.1221483 0.7311733 1.7021922 1.2050789
## DMCC5009144 DMCC6418065 DMCC6484785 DMCC6661074 DMCC6671683 DMCC6721369
## 0.8460941 0.9366789 1.7051477 1.3783664 0.9051912 0.9766868
## DMCC7297690 DMCC7921988 DMCC8033964 DMCC8050964 DMCC9441378 DMCC9478705
## 1.3092629 2.0403024 0.8997665 1.6099094 2.4952706 1.5766959
## DMCC9953810
## 1.5094231
## Fixed effects: rt ~ trial.type * run
## Value Std.Error DF t-value p-value
## (Intercept) 787.5277 18.064731 10089 43.59476 0.0000
## trial.typeincon 75.4953 9.052337 10089 8.33986 0.0000
## run 22.3310 7.293891 10089 3.06160 0.0022
## trial.typeincon:run 2.9946 5.016553 10089 0.59695 0.5506
## Correlation:
## (Intr) trl.ty run
## trial.typeincon -0.226
## run 0.047 0.318
## trial.typeincon:run 0.103 -0.725 -0.200
##
## Standardized Within-Group Residuals:
## Min Q1 Med Q3 Max
## -3.8309061 -0.6703863 -0.1488609 0.4980128 4.6330611
##
## Number of Observations: 10141
## Number of Groups: 49## get unconditional / marginal covariances
m <- rbind(c(0, 1, 0, 0), c(0, 1, 0, 1)) ## contrast matrix
tau.trim <- getVarCov(fit1.het.run.trim)
(tau.trim <- m %*% tau.trim %*% t(m)) ## covariance matrix## [,1] [,2]
## [1,] 1119.643 1150.952
## [2,] 1150.952 1305.353(r.marginal.trim <- cov2cor(tau.trim)[1, 2]) ## correlation## [1] 0.952036svd(getVarCov(fit1.het.trim), nv = 0L)$d ## covmat not degenerate (eigvals > 0)## [1] 22654.493 1429.6471.2 error analysis
1.2.1 test variance in stroop effect
behav <- behav %>% mutate(error = 1 - acc)
fit.error0 <- glmer(
error ~ trial.type + (1 | subj),
behav %>% filter(response.final != "unintelligible", is.analysis.group),
family = binomial,
control = glmerControl(optimizer = "bobyqa", optCtrl = list(maxfun = 1E9))
)
fit.error1 <- glmer(
error ~ trial.type + (trial.type | subj),
behav %>% filter(response.final != "unintelligible", is.analysis.group),
family = binomial,
control = glmerControl(optimizer = "bobyqa", optCtrl = list(maxfun = 1E9))
)
(er.stroopvar <- anova(fit.error0, fit.error1))## Data: behav %>% filter(response.final != "unintelligible", is.analysis.group)
## Models:
## fit.error0: error ~ trial.type + (1 | subj)
## fit.error1: error ~ trial.type + (trial.type | subj)
## Df AIC BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## fit.error0 3 1729.8 1751.6 -861.89 1723.8
## fit.error1 5 1733.0 1769.3 -861.52 1723.0 0.7416 2 0.69021.2.2 estimate cross-run reliability
fit.error1.run <- glmer(
error ~ trial.type * run + (trial.type * run | subj),
behav %>% filter(response.final != "unintelligible", is.analysis.group),
family = binomial,
control = glmerControl(optimizer = "Nelder_Mead", optCtrl = list(maxfun = 1E9))
)## Warning in checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv, :
## Model failed to converge with max|grad| = 0.047928 (tol = 0.001, component 1)summary(fit.error1.run)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: error ~ trial.type * run + (trial.type * run | subj)
## Data:
## behav %>% filter(response.final != "unintelligible", is.analysis.group)
## Control:
## glmerControl(optimizer = "Nelder_Mead", optCtrl = list(maxfun = 1e+09))
##
## AIC BIC logLik deviance df.resid
## 1740.3 1841.9 -856.1 1712.3 10516
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -0.4186 -0.1392 -0.0843 -0.0528 19.4999
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## subj (Intercept) 0.3950 0.6285
## trial.typeincon 1.9923 1.4115 0.75
## run 0.2797 0.5288 0.85 0.43
## trial.typeincon:run 0.8166 0.9036 -0.81 -0.95 -0.63
## Number of obs: 10530, groups: subj, 49
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -5.2767 1.0806 -4.883 1.04e-06 ***
## trial.typeincon -0.1373 1.2254 -0.112 0.911
## run -0.2723 0.7203 -0.378 0.705
## trial.typeincon:run 0.8742 0.7929 1.102 0.270
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) trl.ty run
## tril.typncn -0.858
## run -0.923 0.823
## trl.typncn: 0.828 -0.943 -0.904
## convergence code: 0
## Model failed to converge with max|grad| = 0.047928 (tol = 0.001, component 1)plot(rePCA(fit.error1.run)$subj)
1.3 extract blups and draw figures
## get RT blups
blups <- as.data.frame(coef(fit1.het.trim))
blups %<>% rename(congr = "(Intercept)", stroop = "trial.typeincon") %>% tibble::rownames_to_column("subj")
## bind with error blups
blups %<>%
full_join(
data.frame(
subj = rownames(coef(fit.error1)$subj),
er.logit.stroop = coef(fit.error1)$subj$trial.typei, ## extract logits
er.logit.congr = coef(fit.error1)$subj[["(Intercept)"]]
) %>%
mutate(
er.logit.incon = er.logit.stroop + er.logit.congr, ## logit of error on incon trials
## blup stroop effect in units percent error:
stroop.er = (logit2prob(er.logit.incon) - logit2prob(er.logit.congr)) * 100
) %>%
dplyr::select(subj, stroop.er),
by = "subj"
)## Warning: Column `subj` joining character vector and factor, coercing into
## character vector## draw figure
plot.behav <-
blups %>%
mutate(subj = factor(subj, levels = subj[order(stroop, decreasing = TRUE)])) %>%
dplyr::select(subj, stroop, stroop.er) %>%
reshape2::melt() %>%
filter(!is.na(value)) %>%
ggplot(aes(subj, value)) +
facet_grid(
rows = vars(variable), scales = "free", switch = "y",
labeller = as_labeller(c(stroop = "Response time", stroop.er = "% error"))
) +
geom_segment(aes(x = subj, y = 0, xend = subj, yend = value), color = "grey50", size = geom.line.size) +
geom_point(fill = "grey30", color = "black", shape = 21, size = geom.point.size) +
coord_capped_cart(left = "both") +
xlab("Subject") +
theme(
panel.grid = element_blank(),
panel.border = element_blank(),
panel.background = element_blank(),
strip.placement = "outside",
strip.background = element_blank(),
strip.text = element_text(size = axis.title.size),
axis.line.y = element_line(size = axis.line.size*0.6),
axis.text.y = element_text(size = axis.text.size),
axis.text.x = element_blank(),
axis.ticks.y = element_line(size = axis.line.size),
axis.ticks.x = element_blank(),
axis.title = element_text(size = axis.title.size),
axis.title.y = element_blank()
)## Using subj as id variablesggsave(here("out", "behav", "stroop-blups.pdf"), height = 2.5, width = figwidth)
## write
write.csv(blups, here("out", "behav", "stroop_blups_rt_group201902.csv"), row.names = FALSE)
write.csv(behav, here("out", "behav", "behavior-and-events_group201902_with-subset-cols.csv"), row.names = FALSE)
behav.mod.objs <- list(
fit1.het.trim = fit1.het.trim,
er.stroopvar = er.stroopvar,
rt.stroopvar = rt.stroopvar,
rt.hom.v.het = rt.hom.v.het,
r.marginal.trim = r.marginal.trim
)
saveRDS(behav.mod.objs, here("out", "behav", "mod_objs.RDS"))2 microphone comparison
2.1 read data
2.2 preliminary look
behav %>%
mutate(subj = factor(as.factor(subj), levels = micinfo[order(micinfo$mic), "subj"])) %>%
ggplot(aes(subj, rt, color = mic)) +
geom_point(position = position_jitter(width = 0.1), size = 0.5, alpha = 0.3) +
geom_boxplot(position = position_nudge(1/3), notch = TRUE, width = 0.2) +
scale_color_brewer(type = "qual", palette = 2) +
theme(
axis.text.x = element_text(angle = 90, hjust = 0, vjust = 0),
legend.position = c(0.8, 0.8)
)## Warning: Removed 110 rows containing non-finite values (stat_boxplot).## Warning: Removed 110 rows containing missing values (geom_point).
behav %>%
group_by(mic, subj) %>%
summarize(freq = sum(rt == 0, na.rm = TRUE)) %>%
ggplot(aes(mic, freq)) +
geom_point(size = 2) +
labs(y = "frequency of rt == 0 per subj")
behav %>%
group_by(mic, subj, error) %>%
summarize(freq = sum(error, na.rm = TRUE)) %>%
ggplot(aes(mic, freq)) +
geom_point(size = 2, position = position_jitter(width = 0.1), alpha = 0.4) +
labs(y = "frequency of errors per subj")
behav %>%
group_by(mic, subj, acc.final) %>%
summarize(freq = n()) %>%
ggplot(aes(mic, freq)) +
facet_grid(cols = vars(acc.final)) +
geom_point(size = 2, position = position_jitter(width = 0.1), alpha = 0.4) +
labs(y = "frequency of acc.final codings per subj")
behav %>%
ggplot(aes(sample = rt, color = mic)) +
stat_qq(alpha = 0.3, size = 0.15) +
stat_qq_line(size = 0.25) +
facet_wrap(vars(subj)) +
scale_color_brewer(type = "qual", palette = 2)## Warning: Removed 110 rows containing non-finite values (stat_qq).## Warning: Removed 110 rows containing non-finite values (stat_qq_line).
behav.mean.sd <- behav %>%
group_by(mic, subj, session) %>%
summarize(rt.mean = mean(rt, na.rm = TRUE), rt.sd = sd(rt, na.rm = TRUE))
grid.arrange(
behav.mean.sd %>%
ggplot(aes(mic, rt.mean)) +
facet_grid(vars(session)) +
geom_point(position = position_jitter(width = 0.1)) +
geom_boxplot(position = position_nudge(1/3), notch = FALSE, width = 0.2),
behav.mean.sd %>%
ggplot(aes(mic, rt.sd)) +
facet_grid(vars(session)) +
geom_point(position = position_jitter(width = 0.1)) +
geom_boxplot(position = position_nudge(1/3), notch = FALSE, width = 0.2)
)
2.3 model
2.3.1 differences in mean stroop effect btw mics?
- leave in validation-set data for additional power
behav.rt.mic <- behav %>% filter(!is.implausible.rt, !is.artifactual.rt, acc == 1, mic != "unknown")
fit <- lmer(
rt ~ trial.type * mic + (trial.type | subj),
behav.rt.mic
)
summary(fit)## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: rt ~ trial.type * mic + (trial.type | subj)
## Data: behav.rt.mic
##
## REML criterion at convergence: 156411.5
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.1439 -0.5426 -0.1402 0.3388 10.7989
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## subj (Intercept) 12992 113.98
## trial.typeincon 1233 35.11 0.29
## Residual 27684 166.38
## Number of obs: 11948, groups: subj, 57
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 718.81 33.39 54.93 21.528 < 2e-16 ***
## trial.typeincon 91.14 12.29 53.67 7.413 9.09e-10 ***
## micmicro 104.34 37.58 54.95 2.776 0.0075 **
## trial.typeincon:micmicro -12.70 13.85 53.84 -0.917 0.3633
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) trl.ty micmcr
## tril.typncn 0.153
## micmicro -0.888 -0.136
## trl.typncn: -0.136 -0.888 0.153fit.het <- lme(
rt ~ trial.type * mic,
random = ~ trial.type | subj,
weights = varIdent(form = ~ 1 | subj),
data = behav.rt.mic,
control = cl1
)
summary(fit.het)## Linear mixed-effects model fit by REML
## Data: behav.rt.mic
## AIC BIC logLik
## 152783.6 153256.4 -76327.81
##
## Random effects:
## Formula: ~trial.type | subj
## Structure: General positive-definite, Log-Cholesky parametrization
## StdDev Corr
## (Intercept) 114.40479 (Intr)
## trial.typeincon 32.88231 0.289
## Residual 158.98579
##
## Variance function:
## Structure: Different standard deviations per stratum
## Formula: ~1 | subj
## Parameter estimates:
## 115825 123117 130114 130518 132017 138837
## 1.0000000 1.3680243 0.9491809 0.7273882 1.1413267 0.9937305
## 141422 150423 158136 165032 171330 178243
## 1.0374857 1.3434624 0.6237094 1.6795750 0.9487168 1.5708969
## 178647 178950 197449 203418 204319 233326
## 1.1122717 0.7281854 1.2069523 0.7358722 1.6734130 2.0684924
## 300618 317332 346945 352738 393550 448347
## 0.5321997 0.6442652 0.8708645 1.6949101 0.7676516 1.1962199
## 580650 594156 601127 765864 814649 843151
## 2.0987369 0.6359272 1.1860198 0.5632435 0.9543656 1.2030346
## 849971 873968 877168 DMCC1328342 DMCC1596165 DMCC1971064
## 1.5060506 0.7877884 1.3172820 0.3994237 1.6274925 0.8102444
## DMCC2442951 DMCC2803654 DMCC2834766 DMCC3062542 DMCC4191255 DMCC4854075
## 0.4411251 0.6736451 0.6987490 0.9810488 0.6522941 0.6798883
## DMCC5009144 DMCC5195268 DMCC5775387 DMCC6371570 DMCC6418065 DMCC6661074
## 0.5090925 0.5882508 1.4021924 0.9881116 0.5390242 0.7096957
## DMCC6671683 DMCC6705371 DMCC6721369 DMCC7297690 DMCC8033964 DMCC8050964
## 0.5231536 0.7451889 0.5573742 0.7791344 0.4987172 0.8518369
## DMCC9441378 DMCC9478705 DMCC9953810
## 1.3638986 0.8554003 0.9115365
## Fixed effects: rt ~ trial.type * mic
## Value Std.Error DF t-value p-value
## (Intercept) 721.5696 33.38987 11889 21.610437 0.0000
## trial.typeincon 86.4733 11.16612 11889 7.744258 0.0000
## micmicro 100.3302 37.59825 55 2.668481 0.0100
## trial.typeincon:micmicro -7.0790 12.63690 11889 -0.560188 0.5754
## Correlation:
## (Intr) trl.ty micmcr
## trial.typeincon 0.182
## micmicro -0.888 -0.162
## trial.typeincon:micmicro -0.161 -0.884 0.178
##
## Standardized Within-Group Residuals:
## Min Q1 Med Q3 Max
## -4.8864061 -0.6227512 -0.1716170 0.4302583 10.0354679
##
## Number of Observations: 11948
## Number of Groups: 57- fomri microphone recorded faster RTs
- no observed impact of microphone on size of mean stroop effect ### differences in variance between mics?
fit.het.ml <- update(fit.het, . ~ ., method = "ML")
fit.het.mic.ml <- update(fit.het.ml, . ~ ., weights = varIdent(form = ~ 1 | mic))
fit.hom.mic.ml <- update(fit.het.ml, . ~ ., weights = NULL)
(anova.mic <- anova(fit.het.ml, fit.het.mic.ml, fit.hom.mic.ml))## Model df AIC BIC logLik Test L.Ratio p-value
## fit.het.ml 1 64 152811.9 153284.8 -76341.96
## fit.het.mic.ml 2 9 156422.1 156488.6 -78202.06 1 vs 2 3720.196 <.0001
## fit.hom.mic.ml 3 8 156456.2 156515.3 -78220.10 2 vs 3 36.081 <.0001modobjs <- list(
anova.mic = anova.mic,
mic.model.means = fit.het
)
saveRDS(modobjs, here("out", "behav", "mod_objs_mic.RDS"))- microoptics microphone recorded more variable RTs
3 validation-set subjs: estimation of stroop effects
3.1 read data
3.2 model, plot, and write
## initial fit
behav.vset.rt <- behav %>% filter(acc == 1, !is.na(rt), rt < 3000, rt > 250, !is.analysis.group)
fit.vset <- lme(
rt ~ trial.type,
random = ~ trial.type | subj,
data = behav.vset.rt,
weights = varIdent(form = ~ 1 | subj),
control = lmeControl(maxIter = 1e5, msMaxIter = 1e5, niterEM = 1e5, msMaxEval = 1e5),
method = "REML"
)
## trim and re-fit
behav.vset.rt$resid.p <- resid(fit.vset, type = "p")
behav.vset.rt$is.far.out <- farout(behav.vset.rt$resid.p)
fit.vset.trim <- update(fit.vset, subset = !is.far.out)
## model error
fit.error.vset <- glmer(
error ~ trial.type + (trial.type | subj),
behav %>% filter(response.final != "unintelligible", !is.analysis.group) %>% mutate(error = 1 - acc),
family = binomial,
control = glmerControl(optimizer = "bobyqa", optCtrl = list(maxfun = 1E9))
)## boundary (singular) fit: see ?isSingularsummary(fit.error.vset)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: error ~ trial.type + (trial.type | subj)
## Data:
## behav %>% filter(response.final != "unintelligible", !is.analysis.group) %>%
## mutate(error = 1 - acc)
## Control: glmerControl(optimizer = "bobyqa", optCtrl = list(maxfun = 1e+09))
##
## AIC BIC logLik deviance df.resid
## 680.3 711.4 -335.2 670.3 3661
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -0.4709 -0.1620 -0.1031 -0.0076 9.6958
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## subj (Intercept) 17.87 4.227
## trial.typeincon 10.12 3.181 -1.00
## Number of obs: 3666, groups: subj, 17
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -11.961 4.715 -2.537 0.0112 *
## trial.typeincon 7.983 4.657 1.714 0.0865 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## tril.typncn -0.998
## convergence code: 0
## boundary (singular) fit: see ?isSingular## extract predictions
blups.vset <- as.data.frame(coef(fit.vset.trim))
blups.vset %<>% rename(congr = "(Intercept)", stroop = "trial.typeincon") %>% tibble::rownames_to_column("subj")
## bind with error blups
blups.vset %<>%
full_join(
data.frame(
subj = rownames(coef(fit.error.vset)$subj),
er.logit.stroop = coef(fit.error.vset)$subj$trial.typei, ## extract logits
er.logit.congr = coef(fit.error.vset)$subj[["(Intercept)"]]
) %>%
mutate(
er.logit.incon = er.logit.stroop + er.logit.congr, ## logit of error on incon trials
## blup stroop effect in units percent error:
stroop.er = (logit2prob(er.logit.incon) - logit2prob(er.logit.congr)) * 100
) %>%
dplyr::select(subj, stroop.er),
by = "subj"
)## Warning: Column `subj` joining character vector and factor, coercing into
## character vector## plot
plot.behav.vset <-
blups.vset %>%
mutate(subj = factor(subj, levels = subj[order(stroop, decreasing = TRUE)])) %>%
dplyr::select(subj, stroop, stroop.er) %>%
reshape2::melt() %>%
filter(!is.na(value)) %>%
ggplot(aes(subj, value)) +
facet_grid(
rows = vars(variable), scales = "free", switch = "y",
labeller = as_labeller(c(stroop = "Response time", stroop.er = "% error"))
) +
geom_segment(aes(x = subj, y = 0, xend = subj, yend = value), color = "grey50", size = geom.line.size) +
geom_point(fill = "grey30", color = "black", shape = 21, size = geom.point.size) +
coord_capped_cart(left = "both") +
xlab("Subject") +
theme(
panel.grid = element_blank(),
panel.border = element_blank(),
panel.background = element_blank(),
strip.placement = "outside",
strip.background = element_blank(),
strip.text = element_text(size = axis.title.size),
axis.line.y = element_line(size = axis.line.size*0.6),
axis.text.y = element_text(size = axis.text.size),
axis.text.x = element_blank(),
axis.ticks.y = element_line(size = axis.line.size),
axis.ticks.x = element_blank(),
axis.title = element_text(size = axis.title.size),
axis.title.y = element_blank()
)## Using subj as id variablesplot.behav.vset
## write and save
ggsave(here("out", "behav", "stroop-blups-validation.pdf"), height = 2.5, width = figwidth/2)
write.csv(blups.vset, here("out", "behav", "stroop_blups_rt_group201902_validation.csv"), row.names = FALSE)
saveRDS(fit.vset.trim, here("out", "behav", "fit1-het-trim_group201902_validation_set.RDS"))